Papers › DeiT-LT Distillation Strikes Back for Vision Transformer Training on Long-Tailed Datasets

DeiT-LT Distillation Strikes Back for Vision Transformer Training on Long-Tailed Datasets

3 Apr 2024arXiv:2404.02900archive 2025-07-28

Harsh Rangwani, Pradipto Mondal, Mayank Mishra, Ashish Ramayee Asokan, R. Venkatesh Babu

Vision Transformer (ViT) has emerged as a prominent architecture for various computer vision tasks. In ViT, we divide the input image into patch tokens and process them through a stack of self attention blocks. However, unlike Convolutional Neural Networks (CNN), ViTs simple architecture has no informative inductive bias (e.g., locality,etc. ). Due to this, ViT requires a large amount of data for pre-training. Various data efficient approaches (DeiT) have been proposed to train ViT on balanced datasets effectively. However, limited literature discusses the use of ViT for datasets with long-tailed imbalances. In this work, we introduce DeiT-LT to tackle the problem of training ViTs from scratch on long-tailed datasets. In DeiT-LT, we introduce an efficient and effective way of distillation from CNN via distillation DIST token by using out-of-distribution images and re-weighting the distillation loss to enhance focus on tail classes. This leads to the learning of local CNN-like features in early ViT blocks, improving generalization for tail classes. Further, to mitigate overfitting, we propose distilling from a flat CNN teacher, which leads to learning low-rank generalizable features for DIST tokens across all ViT blocks. With the proposed DeiT-LT scheme, the distillation DIST token becomes an expert on the tail classes, and the classifier CLS token becomes an expert on the head classes. The experts help to effectively learn features corresponding to both the majority and minority classes using a distinct set of tokens within the same ViT architecture. We show the effectiveness of DeiT-LT for training ViT from scratch on datasets ranging from small-scale CIFAR-10 LT to large-scale iNaturalist-2018.

PaperPDFCode

In Syntology Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

Code

val-iisc/DeiT-LT officialmentioned on GitHubpytorch report
pwc-1/Paper-8 mindspore report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Image ClassificationInductive BiasLong-tail Learning

Results from the paper archive 2025-07-28

TaskDatasetModelMetricValueRank at snapshotLeaderboardReport
Image Classification iNaturalist b_22DeiT-LT(ours) Overall 75.1 #19 of 19 Archive leaderboard report
Long-tail Learning CIFAR-10-LT (ρ=100) DeiT-LT Error Rate 12.5 #5 of 28 Archive leaderboard report
Long-tail Learning CIFAR-10-LT (ρ=50) DeiT-LT Error Rate 10.2 #3 of 8 Archive leaderboard report
Long-tail Learning CIFAR-100-LT (ρ=100) DeiT-LT Error Rate 44.4 #11 of 66 Archive leaderboard report
Long-tail Learning CIFAR-100-LT (ρ=50) DeiT-LT Error Rate 39.5 #8 of 25 Archive leaderboard report
Long-tail Learning ImageNet-LT DeiT-LT Top-1 Accuracy 59.1 #16 of 69 Archive leaderboard report
Long-tail Learning iNaturalist 2018 DeiT-LT Top-1 Accuracy 75.1% #15 of 43 Archive leaderboard report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

Absolute Position EncodingsAdamAttentionBPEDense ConnectionsDropoutFocusLabel SmoothingLayer NormalizationLinear LayerMulti-Head AttentionPosition-Wise Feed-Forward LayerResidual ConnectionSETSoftmaxTransformer

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections